Meet the most AI-pilled CEO on Earth

Support for this show is brought to you by Solonus. Does it ever feel like you're being told to wave a magic AI wand and everything will just get better? Sure, AI can chat and summarize, but what about big business problems, like rerouting stock through a canal that won't unblock? Enterprise AI needs context for that, so Solonus provides it. The Solonus context model gives AI operational clarity, so agents know how your unique business runs and how to prove it. Meet the model at C-E-L-O-N-I-S dot com slash context. This episode is brought to you by Facebook. So you were scrolling on Marketplace, and there it was. The bike you've been searching for, you sent a message, and it turned out the seller was super chatty, kind of funny, and an avid cyclist. The next thing you know, you're in a cycling crew, well a community cycling group. The thing about Facebook, you might find more than what you're looking for. From a browse to a bike ride, this summer find more on Facebook. Before we get into today's show, a quick note. This is the final episode of Access. You can keep following me at sources.news and [email protected]. But don't go anywhere. The show's feed is going to live on, and you're going to be hearing a lot more from me here very soon. Talk more after the show about why we're winding everything down, what we're proud of, and what's next for the both of us. There are some interesting nuances to evaluating models. A model that you don't like on day one can be a trash model. If you use it on low or medium thinking, and you get rid of your skills, it's way better. Today, we have a guest who has been living at the edge of how AI is changing work, media, and the internet. Dan Shipper, CEO of Everie. You have to be willing to waste tokens. I was testing you 5.6, so I'm working the next day to a mess from Ariel, saying, "Did you spend $2 billion tokens overnight?" I was like, fuck. Dan breaks down why he thinks we'll all spend more time inside AI coding agents, how every is building agents to scale human taste, and why the real opportunity is not automating everything. It's figuring out what humans do next. Plus, his two phone system, billion token experiments, and why Siri may become a much bigger threat to ChatGPT than people realize. Welcome to Access. I love your little, every mic cube thing. Is there like an industry term for that that I'm not aware of? I have no idea. This is like, I actually, I was thinking about this because this is the first time I've seen this, and I thought you guys would like this, because it's sort of, it's one of those things where you spend like six and a half years, like pushing a boulder up a hill, and you're like every single thing that happens you did, and then eventually it starts rolling down the hill, and you show up to your recording studio, and one of these is there, and you're like, I don't need to do it this, but I love it, you know, and like that's a really nice feeling. That's when you know you've made it. You got the little cube, and you didn't do the cube. But was it an actual member of the team that did it, or was it a gun for hire hired by one of your agents, who just swooped into the office, added that to your mic and it was obviously Codex. I mean, poke did this the other day, did you guys see this? They had poke humans on July 4th, and you could have them go do task gravity things, but I was too afraid to use that. Poke bought by Cognition, shout out Marvin, former guest, a sentence that only a certain cohort of people will understand Cognition buying, but those are two companies. I think of you and every is the most AI-pilled human and organization out there that isn't a frontier AI lab, and people know you guys for the work you do on model testing. You are putting out a lot of writing as well. I think you're working on a codex history of codex, Opus, although I shouldn't use that word in this context, but a sonic story, a sonic, yeah, about the history of codex, and you've got this interesting hybrid media consulting tech software company that you're building, and we want to dive into all of that. But first, I thought it'd be interesting, actually, because I haven't heard you talk about this super recently. Can you just take us through a normal day for you and how you use AI? I assume it's changing on an almost daily basis, because you're testing everything, but what's your day like? Are we getting out of bed and having an extended voice conversation with chat? What are we doing these days? It is changing a lot, and I'll say it has changed radically in the last day or two, because opening up is new. We've gone from years to months now to days. It doesn't change radically every day, but some days more happens than in weeks or years combined, you know, whatever that quote is. So I'm definitely in one of those phase changes right now, because chat, GBT, work, and codex's voice integration, it's so freaking good. So you're working from the toilet now is what you're saying everywhere. Everywhere is the least of it. I can talk about today, and today is a little bit of a different kind of day, because I explicitly took the week off to write this piece on writing a history of codex. I think codex is one of the most interesting product and just business stories of the last like 10 or 15 years, specifically open AI, synonymous with AI, launch chat, GBT, and then found themselves for the last year kind of behind Anthropic, and I think they use codex to come back as a product which kind of disrupted themselves, and then they merged back in. That like almost never happens, and I think there's a lot of really interesting questions to be explored about how they did that and why it actually worked this time. But I took the week off to write that, and normally basically I structured my day with I spent half my day writing and half my day operating, and as the company has gotten bigger, we're about 30 people now that has become more and more intense, and I'm still shooting for that, but it's a little bit harder to hit, so now I'm also trying adding in, every once in a while, do like a little stacation where I write, and that's been great. So I got to get up today, what did I do when I got up? One of the, okay, this is just all going to be me talking about chat, GBT voice, so sorry, but there's a lot to say here. So one thing that's been really, you turned your dreams into action, five minutes after waking up. One thing that's been really interesting about voice is it now has time awareness, so I use it for doing meditations, so I'm just like, on to do a 30-minute meditation, let's do a guided set of timer, here's the kind of style of meditation I like, and it's actually pretty good. I think that there's some room for improvement, I've been doing it on my phone, and I think if I did it in the chat, GBT for work app, where it would have access to my whole computer and stuff like that, I could have it grab transcripts from meditation teachers I like and add to that, but it'll get there eventually, it's just not all hooked up to everything right now. But anyway, I did one of those, it was great, very, very helpful, especially to do a guided meditation that is personalized for me, and what I'm feeling, and I can talk to it as I'm meditating, and it can interact with me a little bit, which is kind of interest thing, and so I did that, and then I got up, I did a bunch of reading, I'm reading, being in time by Heidegger, and a couple other random things, but in particular, I don't know how much Heidegger you guys have read, but he's fucking impossible to understand. I've done several things to help me with this, the first thing, which has been my main method of reading Heidegger is I vibe coated with fable, an app called Verso, that has all the text because it's out of copyright, and then I read it in paper. Not that that would matter, but yeah, it's out of copyright, I have all the text. I read it in paperback, but it's sort of like a one-shot kind of lap, and I go to whatever page I'm on, and I can turn the page over, and it has a plain English explanation of whatever I'm reading, and then any of the words I can highlight them, it saves all my highlights, it also, I can just press explain, it'll explain it, I can say what is the German, it'll talk about the German, and how the translation like misses things. It also, there's a professor I love, his name is Hubert Dreyfus, who wrote, famously wrote this book about why the first generation of AI wouldn't work called, what computers can't do, and it was all based on Heidegger's philosophy, and really good, I love that guy, and he has a lecture course on being in time that is on archive.org, and as part of this, the fable went in like grab the lecture, all the lectures restored it, put it into a player experience, transcribed it, and then also linked every page of the book to the parts of the lectures where he talks about that page, and so basically as I'm reading, I just like go to the part of my app, and then I flip it over, and then I see what Dreyfus says, and that really helps. But what I've been doing recently, over the last, and recently is like literally like last two days, I can just throw on chat to keep voice, and say, here's where I am in the book, and I just read the book to it, and then I'm like, what the fuck does this mean? And it's very good at helping me understand it, and just like waiting for me, I can read Thank you for 10 of you. 15 minutes and then be like, "I'm stuck here." I wouldn't do that. I'm literally having to explain every single sentence because I'm in a very dense part of the book. But I could if I was smarter, let it just like hang out with me and then talk to me as I read. And I think there's so many places to go here with using it as a companion to do intellectual exploration that doesn't require a screen. But that's the first hour after I get up. I'll stop there. Well, that's interesting because, I mean, they haven't confirmed it. But I do think that is the thesis of the first device that OpenAI is working on with Johnny Ive. It's going to be, to my understanding, this kind of puck-like device that mostly sits on your desk or you take with you around the house for exactly that kind of use case. So it's interesting to hear you say, like you're already doing that with the phone app. I would guess that too. I mean, it's so funny. I feel so bad for Alexa and Siri right now. Like, they suck so hard. And even the new Siri, have you tried it? I have not tried the new Siri and I've heard it's good. So it's hard. - The new Siri's really good. I've got the public beta on my phone. - Okay, yeah. I've not tried the beta, but I've heard good things and that would be great if it was good. But I do think this is a big opportunity for them because yeah, I don't really want to look at a screen and it's good enough that I don't have to do anymore for a lot of things. - Alexa, are you using new Siri? - I am. My two-year-old phone is crumpling a little bit under the weight of the public beta, but I had to know how the new Siri is working. And I've been really pleasantly surprised. I mean, whether it's pulling from local data from my notes, I mean, I've been switching back to Apple stuff as one does in productivity worlds every year or two to see what's going on with the normies, whether it's Apple mail, reminders, notes, obviously. And I feel like this is the real list reason we've had in years to use those products. And I mean, the other day I was doing like a live shoot with an email app called Vec. And I guess that's one of your core competitors. And we were talking about the whole idea they have with their recent campaign about the cold emails or the emails that you might have missed. And I said, find my first email with Rylin. And it found it in like 10 seconds. - Rylin. - And it took a few days to index everything. And I'm not sure exactly how their, you know, rag implementation works, but it, having access to that material, especially for most people who aren't writing MD files of everything, I feel like is gonna be pretty groundbreaking. - The indexing takes roughly a week. Has been my experience and people I've talked to 'cause they're indexing everything on the phone. But that is the power of Apple is they have this context that no one else has. - No integrations required, no connections required, poke, forgets about my notion integration every week. - I think it's a huge deal. And it would be so fun and funny if Apple did the most Apple-y thing in the world, which is to skip all the two or three year like knife fight and then just release the thing that it wants to use, it would be really funny. - And I've been thinking about what the second order implications of this are. If Siri's really good now, and people are even using the Siri app, like they would use chat, what happens to chat and clot, right? That's an interesting question. And I think the jury's out, obviously. - I don't know, but it really depends on how good those things are. And I think generally Apple is gonna have to be super consummary and I think that we're starting to see a bifurcation of AI into extremely consummary and then power knowledge worker use cases that are reminiscent of code or use cases, but like for people who are non-technical. And those are two different customer use cases. And I think right now, open AI, for example, is trying to put that all into one app into chat, DVD for work, which now has chat, work, and codex. And it seems to be working, but it's also a really tall order to make something that anyone can use. Like if you're just, you know, a mom in the Midwest, you have a question versus you're an AI-pilled builder orchestrating like 15 sub-agents. It's not necessarily clear to me that those should be the same app, but that's where, that's what they're trying to do. And I think Apple doesn't really have to do the work for power users thing. - No. Well, this is so silly, but like this occurred to be yesterday is that as far as we know, Siri AI is gonna be free, right? - Yeah, it's totally free. - Which is kind of crazy. You know, I've been paying $20 a month for poke, for, you know, a lot of other services, and that's quite an advantage. - I think it could be like when. - Even Apple Music is paid. - Yeah, I think it could be a lot and analogy you'll appreciate. Like when Instagram introduced stories and it didn't kill Snapchat, but it. - Trigger warning, please. - Yeah, Ellis worked at Snap at the time then. And it didn't kill Snapchat, but it severely curbed future growth. And a lot of people who would have maybe signed up for Snapchat for stories never did. And I think that could happen with the assistant market in Siri. - Well, I think Dan's potentially a good person to talk about this with is that the bifurcation isn't just kind of the productivity stuff for the personal stuff, but even within the personal stuff, I feel like there's quite a big difference between which AI is my muscle memory for asking quick questions and many, many other use cases. And I think all my quick questions have immediately started going to Siri AI for whatever that's worth. - Uh-oh. - You know, I always think we are, have to be a little bit of a leading edge. So if you're using, I'm not beta testing Siri as a Siri. So I have to ask you, what has happened then to your chat GPT andthropic poke use then? Are you still using them and for what? - Yeah, I think the poke usage has gone down for all the quick questions. And what's funny about it, this sounds so stupid, but you know, the user interface does matter in terms of how upstream you are of whatever someone wants to know. And when I was in the car, I was like hacking it like literally sending poke a voice note because it doesn't have voice mode. And I need to try the new chat GPT voice and whatnot, but it's just so easy now to just hold down that button in my car for Siri or for, you know, car play and just ask a question now. Or when I drive by a restaurant, I say add that to my restaurants to try list. - Ooh, on Sunset Boulevard and I used to have to ask poke to edit a notion note to do that. And sometimes it would not work. - Wait, so Siri knows that your next to a restaurant, you just say add that to my list or do you say the name? - Oh, well, that's a whole other story. No, I say the name, but also what it does know is what songs on. So now I'm in the car. I say what's this song about? And it knows. - Even if you're using software, I say, - Or is it just Apple music? - I'm an Apple music guy, so I don't know. But I mean, it's got to be the same what, now playing API or whatever it is that they offer. I bet it works. Or certainly if I'm like looking at something on my screen, I say add this to my notes or add this list of places and it does it. It's quite, quite useful. - That's so amazing. I got it, I guess I got to get this. I do have an older, old-ish iPhone. I think I have an iPhone 15 or something like that. - Wow, so, Dan, you can expense it. - Dan, I thought you lived at the frontier. I do accept the iPhone. All the iPhone generations are the same. Well, now I have my iPhone 15 and then I have an iPhone 13, which is my house phone. And basically, when I get home, I put the iPhone 15, which has all my work stuff on it. And I take the iPhone 13 off the charger that doesn't even have a cell plan. It's just gonna get to Wi-Fi and it just has, you know, chat you with T and clod or whatever. So that no one can get in touch with me. I can't do any, you know, I can't be browsing stuff. I can't be doing this growing. And that's been, it's a big life hack. Highly recommend. - Wow. You sound like a brick phone person. Like, you do one of those at one point. - I have a brick. - This is my latest attempt at a brick-like lifestyle. Isn't this so interesting? So like, we started this. Like, you're one of the most AI-pilled people I know. And you know, at the same time, you just got a brick. Like, this is this interesting dichotomy of where we're at with tech is like, it's very exciting. We're all trying it and yet we all also crave more disconnection from it than maybe we did in years past. Do you feel that? - I've definitely always thought about this or not always. Like, when I was in high school, I probably wasn't thinking about this. But I've definitely been thinking about this for a while, but I do think it's true. I do think as we get deeper into this new technology paradigm, it has become clearer that, at least to me, that the way that my brain operates when I'm using agents or anything on my computer is just different than the way that my brain operates otherwise. And it has become important for me to protect time that my brain is sort of operating differently. Honestly, it feels better, but B, I do better work. I'm more focused when it's to some degree limited and I'm not flipping back and forth. It makes a lot of sense to me that people would do that. - I mean, setting aside the phone though, I feel like you're really on the cutting edge of a lot of these new use cases with AI. And I was trying to picture you reading. And I feel like there's got to be a million thoughts that come to you the whole time. And I was curious how you file those, how you even take notes on your reading. I feel like it's really hard to focus as every potential thing now, every experience now, even going through the redwoods, which some guy was posting about the other day, has become fodder for potential productivity. - That's a great question. - Which I think can be cool, but also quite worrisome. - I've always read paperback books. And the office, which I mean, is filled with. books and it's all just books that I have that don't don't fit my apartment and I don't really like reading digital books because they all have the same feeling like a paperback or something that can hold just like has a certain vibe that it's unique which I love each one is unique and that helps me remember it and make make it its own experience and I've always been one of these people that every every like true nerd is obsessed with organizing their book notes and like taking notes and remember them what they read and whatever so I've always had that thing and I was into Rome when I came out I've had all these different systems that I've written about like one thing I used to do is I used to take a blank sheet of paper and as I was reading I always underline I always have a pen I always have a red pen on me and that's what I used to like underline my books and then what I used to do is I used to take a blank sheet of paper and I would hand write an index as I was reading so anything that was interesting to me I would like write it on index and then I had the blank sheet of paper in the book and a lot of my old books even in the office have those indexes and then I started doing Rome and then I think probably when I hit 30 I was like this is useless like why am I doing this? I exhausted myself on that as well and I have this I have I have a couple of things like I have this note that I started keeping because of Robin Sloan I didn't interview with Robin Sloan and he's heard he was doing this and so I started doing it which is I called it it's my ineffable list and it's anything that just has that little like flavor of just I just like this sentence for whatever reason it's not necessarily facts it's just like interesting sentences it's a writer's notebook it's a commonplace book and so I have that and that's an apple note and I just add to that all time but I do think that one of the reasons I love AI is this is like this is any note takers dream it's any book lovers dream any any time I'm reading I can just say now I can just say that you know chat you be for work hey like save this passage and it just we'll save it and then it'll bring it up I don't think it's good at bringing it up in the right time yet it's I think it's discernment for which things would be interesting to me is actually pretty poor but I do think that will happen and so it's just the best thing in the world and you're right if you do that all the time you're not really it's not the same kind of reading experience and so for me it never sentence you look something up ever sentence you make a highlight every other sentence you send it to a friend you know so I think it's I think it's both it's a you realize when you have this available how many questions you actually have and how much more deeply you can understand something and then you also have to balance that with it takes you out of it it takes you out of the experience and it makes the experience quite different and which experience do you really want and which one is better for your purposes or where you are right now and for something like Heidegger would not be able to read it without this and sometimes I like it's intentionally read it in a I want to read this first see see what I can get out of it and then you know then do the lookups like but normally I'm just like going back and forth because it's just a possible something like someone like James Baldwin like it's just nice to read James Baldwin I'm fascinated by every media strategy your strategy you just did the the opus five review early test and you were kind of negative on it which was interesting and people were noticing that and I think giving you props for saying how you felt and like being honest about I because what I observe and this is maybe like my old journalist hat is like this kind of cottage industry that sprung up of people who test models early are very close with the labs they get pre-release access to things in exchange for like publishing their thoughts like it kind of breeds this like very I don't know like soft kind of media environment and you all I think are like very enthusiastic about the technology and the promises of it and that's true and everything that I see from you guys but I think you got a lot of props for saying like oh like actually there's things about opus five that I don't like and I'm curious like how you think about that being the CEO of a company that also makes software does consulting all these things but has this media appress you literally have I think a person with a title like editor-in-chief right like so you're you're making journalistic work you're writing this codex thing like you were saying earlier how do you think about that and approach that and maybe specifically on the the reviewing models piece well I think any kind of writing and any approach has its pitfalls I think you're right if you're in the early access community and you're one of these early people and you you know people in the labs it it can feel hard to be negative and it can feel who wants to who wants to just generally talk about something that they're not psyched about it's just kind of like it's better if the model's better you know for your for your views and clicks right that's a little bit of the way the the incentive structures are set up I think on the other hand like if you're a more disinterested objective journalist the incentives can sometimes be to just be super critical and and so it's sort of both have their both have their downfalls I think for us the the the perspective that we come from is really we're trying to do this we're trying to use this stuff to do our work and to lead our lives and we just try to talk about what we like and if we don't like something that is a important feedback mechanism for the labs to make more stuff that we like and it is it is challenging to think about how do we express this because I'm you know I'm generally not pleased to be like this model sucks I don't really want to say that but it's I sort of think of it in the same way as if you're if you have a friend and they're making decisions in their life that you think are you know bad somehow your relationship is like sort of afraid it's better for you to have an honest conversation with your friend and we just happen to have that conversation in public because that that makes your relationship better doesn't it doesn't do anyone any long-term good to not be honest about how you feel and so for example with with opus it was pretty clear very early that we were not not liking it and we gave like real feedback to them in the in the process of testing it to be like here the things that are not working and in their view yeah we were we were critical and what we tried to do is explain there are a lot of there are some interesting nuances to evaluating models a model that you don't like on day one can can be a trash model or it can may it can mean that it's a model whose powers are only unlocked in specific circumstances that mean you have to break your workflow and if you have to break your workflow that is itself a big thing but it also means that there's an opportunity but there's usually or sometimes there's an opportunity to learn how to use the model and get good results so one of the things that we we found are really key in class and who's the GM of Cora found is if you use it on low or medium thinking and you get rid of your skills it's way better and so we put that in the review because it's it's a thing that people should know is like you may not like this model but if you use it in a completely alien way you might get good results and I think that that's started to bear out a little bit like just from the vibe over the last couple days and so I think we have a commitment to just being honest and and if we're critical do it in a way that's constructive because what we really just want is AI that works for us and I think that is a good strategy support for this show is brought to you by Salonis I'm sure you've heard a lot of vague promises about AI like it's a magic solution or a miracle cure or a bit of pixie dust and if you just add a little AI suddenly your business gets faster smarter and easier sure AI can write emails summarize documents and come up with a lot of synonyms for magic what happens when it comes to your biggest opportunities or your toughest business challenges most AI tools will tell you that's a great question then they'll pull from public information and hand back a list of generic suggestions that's not much help when demand suddenly spikes and you're trying to reroute inventory through a supply chain bottleneck halfway around the world Salonis gives AI the context it needs to understand how your business actually works and where it can improve because AI isn't magic but when it has the right context the results are pretty remarkable learn more about how the context model gives enterprise AI operational clarity at C E L O N I S dot com slash context support for this show is brought to you by Salonis ever feel like you're being told to wave this big magic AI wand and everything will just be better sure AI can write emails summarize some documents and even turn out a business plan in a few seconds but what about when it comes to AI taking on your big business opportunities your big problems it will tell you that's a great question then scrape the public domain and give you some generic recommendations not helpful when you've got a spike in demand and you're trying to reroute stock through a canal that just won't unblock Salonis gives AI the context it needs to know how your unique business runs and how to improve it it's sadly not a magic wand but it does lead to some pretty enchanting outcomes AI needs context Salonis provides it learn more about how the context model gives enterprise AI operational clarity at C E L O N I S dot com slash context it's so interesting I think writing about products that you use and love you're a lot more informed on them but you You also may not represent the audience that the company wants. wants to be targeting in the future, which I've always found to be a fascinating dynamic. For example, decades of reporters, myself included writing about Twitter, which I still call Twitter, when they were trying to push to work for Normies, whether it was with the algorithm or otherwise, you now see the dynamic just in general between the ways creators or early adopters use platforms and the masses. And yeah, you wonder how they felt after that and they say, "Well, it's not, maybe it's not for you anymore," or something like that. How are the reactions in your DMs and emails? It's a good question. I think that Apple, for example, is the first company where I'm like, "What they do is Siri, I may not be the best person to really pull that out and think about it." I would trust MKBHD with that. He's just so good at it. The comparison of Twitter and journalists is a really interesting one to pull out for me. And I do think, and this is maybe self-serving, that we have a pretty lasting connection to the direction of a certain segment of AI products. In particular, there has been this historical thing in AI where the stuff that developers do with it today are the stuff that knowledge workers are going to do with it in six to 12 months once the models get better. And so I think there's going to be continued to be a lot of relevance for power users who discover new workflows, especially as the models change, that then get commercialized or productized for people who are not going to spend the time thinking around with Open Claw on the weekend. And so I think we will sit there, but there's going to be more lanes that open up as more and more of the economy and society piles in here. And it'll change that landscape of who gets to cover it and what kind of coverage is best for sure. How do you imagine that evolving, even the next couple of years, can you pull that out a little bit more? You have to give the landscape secrets away. Yeah, the landscape and the landscape and how you see the media around all this evolving. To be honest, I don't know about the media side of it. I think the landscape is there's three or four-ish main constituencies. There are power users who are technical. There are knowledge workers who need to use it for their jobs and therefore businesses and teams and stuff like that. And then there's just regular consumers. Those are three big places that need to be served. There are a ton of obvious economic incentives to serve big enterprises and teams. And do the things that those people want. What those people want is generally a function of what builders wanted like 12 months ago. And so if you go direct for enterprise people, I think you end up being behind. But you have to suffer the short-term pain of following people in every crowd or around the every crowd of people who are people who are power users building at the edge in a way that feels like it's maybe not legible to big enterprise customers. But I think will be. So a really simple example is right now it's like it's pretty obvious if you're programming you should be using cloud code or codex and you probably shouldn't be looking at every line of code that was extremely not obvious a year ago even to people inside of the labs. And to us it was like of course is how we work. And I think that will keep that will sort of keep happening. So I think that the knowledge worker and enterprise market, you can see a preview of where that's going to be in 12 months by looking at what the builder market is. At some point it may be that model progress stops happening as quickly in this generation. And in that case, I don't think that the builders will be necessarily as informative as like Koch needs sock to compliance for their for their teams that are already deep into codex and therefore that like those needs are actually more important. But right now that's not the dynamic. It's really easy to let the tail wag the dog and think about what big enterprises want and then sort of miss like what they will want which is sort of not right now being defined. And then resolving both of those different constituencies with consumers is instead of a single vision is really it's really hard. And I see some some interesting movements so like serious one but you know another company I invested in is called Portola and they have this AI alien friend. And let's have Quentin on the last episode there's episode. You guys know great guy and like you mostly by like you know Millennial Gen X Gen Z women which is a very different like my audience is 80% he said moms are fully on board now. There you go. And it's a very different take on AI that seems like and I don't maybe it's sort of agentic but it's like that's not the primary purpose of it. I think that category for example on super under discussed we'll obviously be very very important and is not it's not obvious how it gets merged in with people who reason codex. One of the other things to me that makes it a bit fuzzier is knowing when you're actually being more productive or when you are doing work about the work. And I wonder like how your team you know reconciles that like oh are we spending our day building products or are we spending our day building better systems for building products. And this was the whole room research dilemma. Totally. It's like oh I spend all my day setting up note infrastructure instead of actually doing the work and that's like one of my favorite memes online is that all the dudes on YouTube who talk about notes and journaling are taking notes about taking notes. Totally. And this is I mean it's it's that exact thing just on steroids because it's so much easier and more fun to spend your whole day vibe coding your system for vibe coding than it was making your room. Exactly. You know my favorite thing to cook is anything with lemons because after you cook something with lemons use your hands smell better. You feel cleaner feel good. And I think and when we can contrast that with like garlic or you know raw fish or chicken or whatever and with chicken you're like washing your hands every like 15 seconds you're like this feels horrible. At least for me I have a CD so like I'm convinced I'm going to pull myself in the emergency room every time I make chicken. But anyway proper use of proper system building proper use of this to like to set up a system to help you do your work happens in the context of your work. It does not happen in theory. So usually you're doing something you're running to a problem and you're like I'm running to this problem all the time. Let me let me just do something real quick to to build a system to like help me face this more efficiently in the future. It is not like building castles and sky and theory that you might use eventually. So it happens in the loop in the context of your work. And usually like lemons it like leaves you feeling better after and if you are instead feeling like you're just like an empty shell of a human like pulling the dopamine lever one more time to see if it finally solves your problem that you can't even really define the problem. That's when you have a sense that hey this isn't quite this isn't quite something's not working here. And luckily I don't really have like I think probably everyone on team struggles with this to some degree but I have not really we've not really had real conversations about this because everyone is shipping stuff all the time. So I think that in itself is a good enough bar is like are you shipping all the time? Yes. Whatever you're doing to do that is fine. The conversations we have in having and I don't have a solution for that I think is really important. It's just token spend. I like you know I was testing what model was it. It was I think it was 5.6 I was testing 5.6 and usually we get tokens for free to do the testing but in this case for whatever reason the tokens were not free. And I woke up the next day I do a message from my my sister is our head of operations I'm woke up the next day to a message from Ariel saying did you spend two billion tokens over night? I was like fuck you're like whoops you need to go raise another round. Yeah. Let's get a lunch today boys. And there's a really it's there's a really interesting tension there because you have to be willing to waste tokens. If you're if you're not willing to waste tokens you're not willing to discover new things obviously if you're spending I mean our our token spend is absolutely in the 50 to 100 K month range if not more. That's a lot of money. That's the point there and come on right that's yeah my personal token spend 100 K month and is it really it's 50 K month? No. No. I mean for the company. Yeah. What is my actual token? I had to switch to my to chat to be T. I had to switch chat to be workspaces. I'm pulling up my my codex my probably get a lot of free tokens. So you're maybe not the best I'm at 9.4 billion lifetime tokens with this on just this workspace. And like yesterday let's see. You know it looks like I'm around 250 million tokens a day ish would be my average. I don't know what that is in pricing terms but it's a lot. Yeah. And this is just personal, like we also run apps that consume tokens. So trying to let people experiment and trying to also then reflect afterwards, was this a good use of our tokens? Like, would you do that again? Is a really difficult problem that is very valuable to solve? I assume we will solve it eventually. I don't have an answer other than my current thought is, if you have any run that spends a billion or more tokens, we send you a quiz that asks you questions about what you were building and how it was built. And if you can't answer the questions, you go on a wall of shame. And if you can answer the questions, then you become a token billionaire for the day. What if I just have codex answer the questions? We'll have to make that not a thing. No cheating. And I think that's like a sort of like lighthearted way to make sure that people, it's not that we're banning you from doing this. It's just like really think about it. And if you think about it and it's worth it, great. If not, don't do it. Well, you consult a lot of companies on their AI strategy and individuals. What are you hearing from them about token spend right now? Does it kind of map to what you're going through? I think it maps. I think, again, it's one of those when you are really, I think, far away from the ground level of how things work and what's going on. It's really easy to swing from one extreme to another. And the first extreme was just spend as many tokens as you can and the second extreme, then what really quickly to like token maxing is bad and blah, blah, blah. And it's like, and there's no ROI from AI and all that kind of stuff. I'm not saying our clients are like that. Our clients are very smart. But like the general narrative is that forcing your organization to use a tool they don't understand as much as possible is obviously going to be a waste. And now that AI is powerful and can run for a long period of time, that the waste is a lot higher than it used to be because it used to just be a chat and then our response. And that's not that many tokens. But now I can spend two billion tokens that I've been thinking about it. And CFOs are looking at that bill being like, fuck, this is just this is really terrible. And so it's swinging to the other extreme of, you can't use tokens anymore. We're limiting it severely, blah, blah, which is also the wrong move. My, I think the general thing that we talk about with clients and the things I see be successful are technical people. You should have a $200 a month plan, non-techno people who should have a $20 a month plan. Generally, you should be able to stay within those limits. And you should identify a few of the people in your organization who you consider to be like true AI early adopters and give them a high token budget because what they will do is experiment and find the workflows that the rest of your organization is going to use it in the limits of their plan and have some sort of escalation process of someone's running into limits that need need to be changed. But something like that feels like a reasonable policy. Sounds like you have a lot on your plate and among them is a handful of different pieces of software, Cora, Spiral, Sparkle, Monologue, Proof. I know some of them have just one person on them, but I'm curious how that's going. I mean, there's so many jobs, you know, within building a successful product from engineering to the product marketing to the road map. How is that, how is that going? I think it's going well. And we're in the middle of a sort of change in that strategy. One of the key early insights that we had is it is absolutely possible now to have a single person running running an app end to end and doing really well at it. I think Naveen who runs Monologue is an extremely good example. It's just him. He's got some contractors, but it's mostly just him. And that product is growing really quickly and is a actually competitive with companies that have raised like 70 million dollars or more. And yeah, there's like some support from us, but really he's like mostly doing it by himself. It's kind of crazy. If you take a talented full stack person and just let them rip, they can get a lot further than you think. And if you, as a company, spin up a lot of a lot of those, you can sometimes end up tending to and you have one person on each thing. And each thing is going sort of well, but none of them are necessarily breaking out. And I think that what we need to do is develop a different move to be like, we've identified, we've explored the territory of a bunch of different apps. We've identified one or two that we're like really focusing behind. And maybe some of them can continue with one person. But if we think it's going to, it's really something that we want to like win the market with, we should put more resources behind it. And so I think we've started to add a subsequent move, which is we generally start with one person. And then to the extent it looks like something that we're going to really put the org behind, we built out, we built out a team. So we did that with, we have an agent product called plus one, which is originally really built mostly by Willie, who's our head of platform and maybe one other person. That has turned into, we haven't released this yet, but it's now in beta, we use it all the time internally, just in every agent. It's like an instantiation of every inside of your company that knows all the things that we know that works in the way that we work that anyone can use in Slack. And that we're going to sell this externally. We will sell this externally. Right now it's in beta, but I mean, to the tier question about media companies, I think this is a really interesting extension of like a media company. So that's going really well. And that has a team like it's a real engineering team. They're all using AI. And it's structured differently than our initial bets. I think this, this particular product, it's very obviously core to every and like what we do. And it's very complicated, much more, much more complicated than take your pick of, you know, I mean, model is very complicated product, but it's to some degree, there's, there's more under your control. I don't know. Navine would probably argue with me about that. So, but it seems to require a bigger investment than a single person if we wanted to, you know, Anthropic has a, has a version of this called tag. If I want to give you a tag, it's hard to do it with one person. So you use the word agent for every agent is this taking actions using the every kind of corpus or is it just like a fancy MCP that has all the data of every that I can. Core core core, full core core status can do everything. It's natively built in with component engineering and all the other ways that we work. Wow. So, what are the early use cases for that that you're having? I mean, internally, that guy that has to be kind of weird because it's literally an AI instantiation of your company, but I assume you're testing with some outside partners. Like, what are they using that for? We are starting to test with outside partners, but that's very early and just as a rule, we only really release things that we use ourselves and like ourselves. So we build for ourselves first and then move out. And I think we were one of the first people about a year and a half ago to start using cloud code and Kiran, who I mentioned earlier, really invented this way of working called compound engineering and in compound engineering. And this is something I worked with him a lot on. It's it's different from from regular engineering in that when you in regular engineering, every time you do a piece of work, it makes the next work harder to do because all the systems depend on each other and the code base is bigger and all that kind of stuff. And in compound engineering, you're trying to make the next unit of work easier that to do than last because what you do is after you do a feature, you look at all the things you learned and then you compound that back into your prompts and your agent harness into all these different places so that the next one, you don't make similar mistakes and things are more clear and all that kind of stuff. And that has grown into a really thriving plugin that lots and lots and lots of people use is actually probably like weirdly our biggest, our most scaled software product, even though it's just open source. And I think that that way of working, this is another example of things going from developers to knowledge workers. I think that way of working is going to come to knowledge workers and that a Slack agent is actually an ideal surface for it, specifically the idea of compounding. So like, I do a unit of work and I compound it back into the agents so the agent gets better over time and is better at helping me do the kind of work that I do. And the reason I think it's really interesting in an organizational context means that everybody else in the organization can do the kind of work that I do, which might sound threatening, but it's actually like the most important thing for expert knowledge workers. And I'll give you an example. You mentioned our editor in chief earlier, Kate. Kate is a tremendously good editor and also has, I don't know, probably a team of probably eight to 10 people now and looks at everything that goes out. So all the all the pieces, but also now landing pages, emails, like all that stuff she looks at and has a particular taste for how it all should how it all should look and fit together. As you can imagine, that's a very stressful thing to have to do while you're also managing a whole team. Well, you're talking to someone on this chat. LSU's to oversee all words at Snapchat. So he, there you go. So you know, so you know, you know, Ellis. And for literally like three years, I've been trying to help automate this and the models weren't good enough and they just got good enough. And so what we have is basically I took a corpus of 30,000 of her historical edits. I turned it into a style guide all automatically. And then we have a skill in the every agent that I can just say, Hey, like throw Google Doc in the Slack at every just like copy edit this. And it will go and make suggested changes that she has made pro previously based on her previous work in the style God, we built in the Google Doc. So instead of Kate having-- to like go in and do every document from scratch. She goes in and there's already a bunch of suggested changes that are like, we think this is what you would do. And then she says yes, no, yes, no, and makes her own edits. A, that sort of gets her closer to a pass that she's comfortable with. And sometimes she doesn't have to even review it for like lower priority things. It's just a test on a test landing page. We'll just run the K copy edit and we're done. So she doesn't even have to see it. But what, what happens then is we just compound that back into after every pass, we see what, what she accepted, what she rejected and what we missed. And then it compounds back into the agent and it just gets better. And that will help her scale her taste to the rest of the organization without taking more of her time. And I think that's actually really, really critical for anybody inside of any organization that has any sort of expert knowledge. There's always going to be things that you're turned to by people who need that, that help from you, but also probably shouldn't take your time because you're repeating yourself all the time. And I think that this agent is going to be really good for the, for those kinds of use cases. So it learns from the context that it's in, even if that's outside of every, it's the every agent, but it could really be like the Alex agent at the end of the day. Yes, well, I think hopefully it will have a bunch of different skills in it that each skill is something like something that Alex knows or something that Alex knows that you can use or anyone you can use. Alex, you can't send your agent to do a paid Yahoo dinner. You have to do that personally. Well, I want to do that. Not yet. Not yet. I want to do that stuff. I mean, I think ultimately that hopefully up levels everyone to do more of what they want to do. I mean, that's the promise of all of this, right? We got to get through the busy work first, but of setting it all up. It's really interesting that you're doing that. I mean, I'd be curious to get your thoughts on this, Dan, a conversation I've had with some founders recently that I've been meeting with and it's come up a couple times is they're like, just make an MCP of your brain. All the conversations you're having that you are comfortable sharing publicly and my agent will deliver it to me in a better, more personalized way than you will through one pass of your newsletter. So I'd actually charge more to just have kind of raw token access to your granola, whatever it is, right? And I'm thinking about doing this actually also on your team has helped me come up with some like, you know, early mocks of it. But it kind of is analogous to the, it's more simple, but to the every agent. Right. And I have thought about this a lot with the future of media is a big part of the future of media hyper personalized, agentically delivered media. I think it's a, I think it's a really important place to explore. We are very far away from that in the sense that, you know, I talked earlier about language models discernment. Can it discern what would be interesting to me of what you think? And if I have a go, if I say, here's my situation, if I have a go through all of your granola notes, it's going to come back with some bullshit. That's like, I can sort of see why you would say this, but like, it's not that interesting. So, so a yes, be doing that well is a really hard problem. And it's sort of open whether or not we're progressing, particularly quickly toward that. But I absolutely think it's like part of the future. And it's a big part of the opportunity for people like us. It's like, obviously, once you read some of the stories and you're into them, you kind of want more access to them. And obviously, you only have so many hours in the day, embodying that in something that you can query and talk to you is like really cool and very important. Ellis, you could have a meeting MCP that scales your work with founders where you can have 10X more clients. There you go. That's my ultimate goal. That is interesting, though. And I mean, I know we're running out of time, but I was thinking a lot about your piece you did, Dan, called after automation about how a lot of this automation raises the bar, integrates best practices scale, but what that inevitably creates is room for what's different and what's new and that inherently kind of only comes from people. And I think that's part of it is that whether it's kind of like Kate's copy editor guide or or my own MCP is that it has to change and it has to change as a result of the stuff that you learn, the experiences you have in your life and whether it's art or marketing or otherwise. If any part of the goal is to be new, then it's something that almost I sensibly can't be replaced by AI. And I would certainly like more research time in my day, even to play a video game, which I do consider research. I'm with you. I'm with you. Yeah, I think AI makes yesterday. This is something I wrote about an after automation. It just makes yesterday's competence available to everybody. It's based on training data. So anything that was done yesterday is available to everyone. But today is a different, it's different. It's a slightly different. And if you have non-experts using yesterday's competence to like solve today's problems, it's going to be close. You're going to be able to one shot an app, but it's not going to be actually good, especially because everyone also has the same thing now. And the job, the role of experts is to take that, what is now commoditized, which is the ability to apply yesterday's competence to any problem and use that to actually make solutions that are a good fit for that problem, that particular problem in person, which is the actual valuable thing. And so I think that is the opportunity for experts in this era. Yeah, not just knowing the whole corpus of information about a specific topic, but being on the edge of moving it forward, which is always going to be different. And I mean, marketing has like never been this objective exercise where there is one right answer. The right answer is always moving forward. And that's how you end up with like 30 sites during the vibe code era, whose headline is all what can you build. And a lot of times, when I talk to clients, it's like, there is no one right answer about your best hook. It could be your benefit. It could be talking about your audience. It could be talking about your history. And it's almost like cyclical, like fashion, like a wheel that just keeps moving based on what's new or fresh at the very minimum. And yeah, I don't know, maybe you could program that into an AI, say, hey, if this is today's latest and greatest cycle back to another possible answer in an area that's always subjective. But at least for now, I feel somewhat insulated. You would, you probably can, we'll be able to do that. But even then, you still have the, well, now I have to choose which of those is good. And so to other people, you're sort of moving, you're moving the capability, but there's still this, I make a distinction in that piece between agency and autonomy. So we think of agents as being agentic, but they're actually just, we're actually just talking about autonomy, the ability to like take something that we give it to do and just do it until it's done. And that's very different from agency, which is internally located desires and beliefs and goals and values, which agents don't really have. And until that changes, you can add any capability that you want. And it's still going to end up being something that we direct and control at the end of the day, which I think is probably a good thing. And I think is often missed in all of these discussions about capabilities. Dan, we have to end it, but I do want to end it on a prediction from you. Six months out from now, what do you think about the way we use AI tools and how that will shift? Is there going to be a new way or a new way we're thinking about these tools and these models and their capabilities? I think Claude Anthropic currently has the mandate of heaven. I think Open AI and codex and Chatchy Pee for work will have the mandate of heaven. I don't think it's permanent, like everything goes back and forth, but they're doing something really good over there. I think that we will probably be spending a lot more time in our coding agent orchestration's surface of choice, whether that's Chatchy Pee for work or the Claude desktop app. And in particular, you use the whole internet inside of your code. I knew. That's my big. That's so crazy using the in-app browser of those tools. I think it's going to be a big deal. Are we inside the in-app browser right now? You are. I never need a browser. That's a first. How does that possibly work? Dan, we really appreciate your time. Good chatting with you. Thanks for coming on. Thanks for having me. Thanks Dan. Take care. Before Alice and I get into why we're winding down access and our reflections on the last year of the show, a reminder that you'll be hearing and seeing more of me and the sources universe here very soon. So don't unfollow, don't unsubscribe. And in the meantime, visit sources.news for the very latest. So you were scrolling on marketplace. And there it was. The bike you've been searching for. You sent a message and it turned out the seller was super chatty, kind of funny, and an avid cyclist. This episode is brought to you by Accenture. When you're If your advertising operations fall out of sync, everything else follows. Spotify and Accenture are working together to reinvent the rhythm of ad sales, using automation, analytics and smarter workflows to simplify campaign delivery and access better data across the business. The result? Less time spent on operations, more time connecting brands with the moments and fandoms that matter most. Learn more at Accenture.com/Spotify All right, thank you to Dan Shipper for being the last episode of Access Closed. Access Closed. Access Closed. You've been waiting to say that. Yes. How are you feeling, man? Feeling good. Yeah. We're the one-year wonder. It was a lot of fun. We made hats. We had amazing guests. Yes. We had a party in the notion vestibule. We learned a lot about making content hashtag in the modern era and I'm just happy we got to hang out once. Me too, man. Me too. It's been great. There's really not like a dramatic reason for this. I think we both have been talking a lot about where we're at in our careers and personal lives and shows take time to put together and all sorts of reasons, but really it just came down to, we both feel like for where we're at, it made the most sense to wind the showdown and do our own things and I'm going to have a lot more coming with sources so stay tuned for that, sources.news and you're going to keep crushing with meeting. From the sound of it, you're actually going to let people subscribe to your brain at some point in the future. That may be in the cards, maybe I will as well. I agree with Dan that it's probably too early. The tools don't feel quite ready and also I just don't think enough people are experiencing AI this way for that to really be a product, but I do think it probably will be eventually. If I were an investor, that would be something I'd be looking at. But yeah, man, no, this has been, this has been awesome. I remember sitting down with you at a coffee shop in East L.A. year and a half ago and just like, should we just do a podcast together and you thought about it for like a day and you were a quick yes and then we were off to the races and man, I would hold up, you know, our guest list against any tech podcast guest list, especially a first year show. The kind of guests we've had on in conversations we've gotten to have are just really incredible. I feel really proud of it and it's a catalog that I'm always going to look back on family. Selfishly you get to meet some of your idols, right? That's one of the fun parts about being in content again, then that's a fun part is I go to happy hours and people treat me differently again. I forgot about that. Go to the Figma conference and they're like, oh yeah, you're back in media now, so I can't tell you this to that. I'm like, oh yeah, that is the thing. I'm so used to that, but that was probably new for you. I mean, yeah, that was the interesting part of this is I never left media. You did for a while and getting you back into it. Yeah. I mean, I feel like you've liked it. I feel like you like the limelight. You like to ham it up a little bit, I mean, be honest. Yeah. It's been a lot of fun. Get some press passes that would have cost me and my business some money. But I think, you know, if anything is becoming clear and I think this relates to the conversation with Dan is that we all owe it to ourselves to find the best format for sharing, monetizing, making use of our strengths, you know what I mean? And content is just one of the ways to potentially do that. And no matter what it is, you want it to feel aligned with what you love doing every day. And yes, certainly getting back into content being reminded of the landscape that we now face. And just how hard it is, whether you have a story as a company or an app or something else you're trying to share, all getting squeezed into the same algorithms and the same expectations with content these days. It was really a whirlwind to be thrown back into that. Yeah. A lot learned. You know, the thing that I will take with me the most are the personal reach outs we would get from people who listen. And we had a couple of people even who came to the party in SF who just reached out cold and were fans and wanted to come. And getting those notes and you were better about sending them than I was. But getting those notes, you know, every week really felt validating and it felt like, oh, this is like, even though, you know, we weren't making a show for millions of people. We're making it for a very specific cohort of tech AI insider nerds. Getting that feedback was super cool to me. I'm sure it was for you as well. I think my only regret is we weren't able to get Johnny. That was my bucket list item. So if and when you get Johnny, I'm going to be stowing away. And then I will pop out. Yep. And I will co-interview him with you and sounds good man. Yeah, I'm planning to, you know, these kind of interviews have been part of my thing and what I've have always done and I'm planning to continue them under the sources umbrella. So more to come on that. And yeah, man, I mean, it's been cool to see you connect with like, guess we've had on the show that then, you know, become people you're working with at meaning. I mean, there's just been a beautiful kind of serendipity to that and seeing kind of how your business has grown over the last year as we've been doing this has been really cool to see. Yeah, that reminds me. I can't use my favorite, my favorite line anymore when I'm talking obviously any advice I give to clients about what's most interesting is my opinion, right? It's like what's most interesting is not objective. But if they really push you back and they say, oh, that's not interesting, say, well, just as one example of you are on my podcast, this is what I'd want to talk about. That was always like a super, a secret superpower I could pull out whenever, whenever was needed. So yeah, I'm going to have to come up with something else. Yeah, you could still do it, just like a theoretical podcast or maybe you do a meaning podcast one day. Who knows? Maybe the LS AI doesn't. Yeah. Yeah, I was considering it is definitely a lot of feedback from friends and fans about wanting something more in the storytelling world. Definitely seems like there's a white space for that. But yeah, I mean, it's also just learning about three years into my company, what I want my life to look like. And yeah, certainly adding one more dimension of founders with tough schedules to work around to my life definitely gave me a few more gray hairs than otherwise. The scheduling behind this stuff is harder than it appears. Yes. What was it like for you kind of being off the news cycle, you know, with these interviews, getting more into like more lifestyle? Yeah. You know, we always talked about that being a core thing we wanted to do. And I'm glad we did and you know, it's, it's been very good. It's been a good experience, it's helped me lean into parts of myself and my intuition and my curiosity that felt a little just kind of inherently closed off by the nature of being a journalist in a newsroom before doing sources and going independent. And really like challenging my assumptions of what journalism can be. This wild west of content creation that we're in that we talked about with Dan, it's kind of fitting I think that we ended with him talking about new media because this has felt like this whole show has felt like an experiment in that. And yeah, realizing that I can bring kind of my journalistic sensibilities to an environment that is also about a good hang and getting to know the person and you've been great at that helping pull that out and even even though I've like tolerated your, your Tyler dank jokes, you have brought a sense of brevity to the podcast so I appreciate that. Not as much brevity as Dan you'll show you up in a hot tub. That was definitely a podcast highlight was having a guest videoing in from a hot tub, but we've had some good funny moments. For those who don't know, we literally pulled together the brand and the whole concept of the show to align with Alex's zuck exclusive. That's right. On day one and we pulled this thing together, top to bottom and what, like three weeks or something. It's crazy. Yeah. I mean, shout out to the homies of Lithuania. Yeah, practica with the K, very, very cool dudes who I saw. It's funny that I think they did a recent poke branding exercise. Of course it did. Yeah, they found the way to the start up clients that we talk about all the time, which is cool. Yeah. Any other highlights for you, for we wrap this? Just getting this selfishly asked for product changes with the founders who make things that I like. That was the main thing I missed from being a reporter. That's the best. Hey, hey, you want to talk to me now, right? And now you have to hear my feedback and my feature requests. It's the best. Yeah. I love doing that with Sam from granola, Ivan from notion. Steve from Reddit. Yeah. it is a special perk of this job. - Well, where can our viewers find you going forward? - Sources.news is gonna be the home for everything going forward, big things coming, so stay tuned. But yeah, sources.news, what about you? - Yeah, I hate to say, I feel like Twitter is just (laughs) been my entire career. I have had a hamburger this year. I've got it hamburger, you could find me on Twitter. I especially can be found. Now that the new algorithmic update, which hopefully sticks around, actually allows my followers to see what I'm posting, whether it is smashing or busted, I feel like when you follow somebody online, you wanna see the sharpest stuff and the not so sharp stuff. Even like the new Strokes album, which I'm obsessed with, it's not their best album ever, but it's always interesting, 'cause who's behind it and what they're trying to do? And I've been really liking seeing a lot more conversations with people's followers as opposed to just like, what's most viral online these days. That was always such a strength of X in being kind of the water cooler for people who wanna talk and think about technology all day. So yeah, you could still find me there. And at meaning.company, I'm currently doing a, I'm switching from universe, which it's not clear if it still exists and is being maintained to Framer. And I was screwing around with Framer and I'm like, oh, I could actually make my own app here. I added a nav bar to my website for the first time that looks like exactly like a liquid glass thing you might find in a cool app these days. And so yeah, look forward to a meaning.company refresh. - All right. - A full brand experience. - Well, I guess I'll read us out here one last time. That is it for this week's show. Thanks to Dan Shipper for being our final access guests. You can find him at every.co. And we really appreciate him coming on. And you can find me as I was saying at sources.news online. Stay tuned for much more. - Access is part of the Vox Media Podcast network. Special thanks to our friends at Hooked Creators. - Yes. - For being such wonderful producers, production partners, thought partners. And thanks most of all to you all for listening. - Yes. Thank you guys, really appreciate it. - All right. - We'll see you on the internet. Bye-bye. - Bye. - Support for the show is brought to you by Salonis. I ever feel like you're being promised AI that will magically solve any problem your business might have. Sure, AI can chat and summarize. But what about big business issues? The ones affecting your unique company? You need the Salonis context model, which gives AI operational clarity. So agents can reason correctly, decide sensibly, and act reliably. AI needs context. Salonis provides it. Meet the model at C-E-L-O-N-I-S. dot com slash context. That's nice. Go on, book it. It's easy. Booking.com, booking.year.